Block/Trace Length

• Definition:
  This field means different things for the three Decoding Styles. In Tailing decoding, it is simply the block length. In Block and Continuous decoding, it is the number of trellis sections that the path must be traced back through before decoding can be done.
  
  • In Tailing decoding, you want the block/trace length to be much larger than the number of zeros that are needed to zero out the encoder at the end of each block. Otherwise, you won't have a decent data rate.
    
  • In Block and Continuous decoding, the trace length should be at least as long as 4 times the number of delay elements in the encoder. This length of trellis sections is the buffer we give the algorithm in order for the paths to converge. This is not a fixed number for all codes or even for all message sequences. In fact, there is nothing that proves that the paths will converge in a finite number of steps. If the code is not very good, you can have a situation where the trace paths never converge. Instead, this number, (4 times the number of delays), is just a guideline. It says that the paths will converge with high probability within this many trellis sections.
    
  • What is meant by converging paths?
    After the forward calculation of the path measures, every state has one surviving branch terminating at it. That same state can then possibly have no surviving branches leaving it. In this case, we know that this state is the end of a path, and if this state is not at the beginning of the trace back this is a dead path. Dead meaning it does not continue forward, and no paths started further down the trellis could possibly connect to this path.  On the other hand, this intermediate state could have more than one surviving branch leaving it. This is highly probable for states on a path that has good path measures since paths with good path measures are promoted by the selection of the best branches from the best states.  This ability to have more than one surviving branch leaving a state but only one branch terminating in the state produces a funneling action in the reverse direction. This is what tends to lead all the possible paths to converge to one path given a sufficient number of steps. This converging phenomenon can be illustrated by turning on the Show Trace Convergence option in the options dialog.
    
• Usage:
  Experiment with different values for the block/trace length to see the consequences of choosing too short a trace length. If the trace length is too short and the paths do not converge within this length, then it is ambiguous which path we should decode from. As an approximation in this situation, the path that started the trace at the best valued state is the one that we decode.
  
• Valid Input:
  The block/trace length should be a positive integer. The program will warn you if you enter a bad length.
  
• Examples:
  For a four state code(2 delay elements in the encoder) with block or continuous style decoding, a good value for the trace length is 8.
  Likewise, for a 8 state code, a good trace length is 12, and for a 16 state code, a good trace length is 16.